In the new UMTS architecture with a connectivity layer and a call control layer, Media Gateways (MG) will be used for the connectivity layer. In the call control layer, Media Gateway Controllers (MGC) will be used. The MGC controls the distribution of new calls to MG. Congestion situations are always troublesome.
A standard mechanism for general congestion situations is based on that the node is associated with a congestion level (CL), being a measure of the present overload of the node. Within each CL, there are a number of sublevels, congestion sublevels (CSL) available for making a smooth transition between conditions relevant for the different main CLs. Each message is assigned a congestion priority (CP) set by the system and depending on the nature of the call request. Such a procedure is recommended in SCCP, ITU-T recommendation Q.714.
A typical way to proceed is:                A message that has a CP larger than the CL is accepted and sent to the node.        A message that has a CP smaller than the CL is blocked and discarded and is thus not sent to the congested node.        From the messages having a CP equal to CL, a first fraction is accepted and sent to the node. The rest is blocked and discarded. This first fraction is equal to 1 minus the ratio of the CSL and the total number of available sublevels.        
This procedure will limit the signalling in case of congestion, and the sublevels will smoothen the transition between the different CLs.
A problem arises when there are more than one link/node/MG to send to, i.e. when the messages that are discarded instead could be redistributed to another exchangeable node instead. In most message systems, this is not at all interesting, since the messages are addressed to one specific node, and no other node is of interest. However, it could be of interest on a higher level than individual messages, e.g. where it is possible to redistribute e.g. a whole series of messages to another node, when the series is started up.
A typical example of such a situation occurs in telecommunication systems, where a call can be considered as a series of messages, started-up by some initialization message or call setup. For instance, in the UMTS architecture, there might be several MGCs and several MGs. Each MGC may select a MG among a set of MGs to be used for a call setup. The reminder of the traffic associated with that call will then continue to use the same MG.
A problem thus arises when going from a one-to-one node system to a one-to-multi or multi-to-multi node system. Congestions at the different nodes have to be handled and an efficient redistribution of load has to be performed.
One possible solution is to always select the least loaded MG. That is the one with the lowest CL (and CSL). In, for instance, telecommunication systems, there are, however, problems connected to this. One problem is that with such a solution, all load within a certain time period will be directed to one and the same node (before the congestion measurement reacts). Since the main load of a call is noticed in the node with a certain time delay, the CL and CSL will start to increase only after some time. When the congestion level of this node passes the congestion level of another node, then all load will be directed to this other node instead. This procedure will subsequently lead to oscillations in the load between the different nodes. With large inertia in the system, high amplitude oscillations and risk for overloading will occur. The problem gets worse and worse with an increasing number of available nodes.